IKEHARA-TYPE THEOREM INVOLVING BOUNDEDNESS JACOB KOREVAAR Abstract. Let ^ a.n/n'^ be a Dirichlct series with nonnegative coef- ficients that converges to a sum function f[z) = f{x + iy) for a; > 1. Setting sn = '^n<N '^n, the paper gives a necessary and sufficient condi- OO , tion for boundedness oi sn/N. As a; \ 1, the quotient f{x + iy)/{x + iy) must converge to a pseudonieasure g(l + iy), the distributional Fourier transform of a bounded function. The paper also gives an optimal esti- mate for sn/N under the 'real condition' f{x) = 0(1/(1 — x)}. o o 1. Introduction We recall the famous Tauberian theorem of Ikehara: Theorem 1.1. Suppose that the Dirichlet series oo en H (1.1) y^ — ^ with ttn > and z = x + iy n=l converges throughout the half-plane {x > 1}, so that the sum function f{z) >• ' is analytic there. Suppose furthermore that there is a constant A such that ^ ■ the difference O '■ (1-2) 9{z) = f{z) - -^ • ' z — i Q ■ has an analytic or continuous extension to the closed half-plane {x > 1}. OO , Then the partial sums Sn = '^n<N ^n satisfy the limit relation i;. ; (1.3) sn/N -* a as N -^oo. ^ . The theorem is often called 'Wiener-Ikehara theorem' because Ikehara H I studied with Wiener, and apphed the new Tauberian method that Wiener was developing in the years 1926-1931; see [1], [TO], [TT] and cf. [5], [6]. Ikehara's theorem led to a greatly simplified proof of the prime number theorem. Date: July 3, 2008. 2000 Mathematics Subject Classification. 40E05. 1 2 JACOB KOREVAAR In [7j the author obtained a two-way form of the theorem. Given f{z) of the form fll.ll) . the following condition is necessary and sufficient for (11.31) : The difference g{z) = g{x + iy) must have a distributional limit g{l + iy) for X \ 1 which is locally equal to a pseudof unction. That is, on every finite interval {—B,B), the distribution g{l + iy) must be equal to a pseudo- function which may depend on B. A pseudof unction is the distributional Fourier transform of a bounded function that tends to at infinity. It can also be characterized as a tempered distribution which is locally given by Fourier series with coefficients that tend to 0. A pseudof unction may have nonintegrable singularities, but not as strong as first-order poles. In connection with Ikehara's theorem one may ask (cf. Mhaskar [8]) what condition on f{z) would suffice for the conclusion that (1.4) sn/N = C(l) as A^ -^ oo. Unlike the situation in the case of power series Yl (^nZ^i it is not enough when /(■) satisfies the 'real condition' (1.5) /(x) = C{l/(x-l)} asx\l. Proposition 1.2. For Dirichlet series (li.ip with sum f{z), condition (I J . 51) implies the estimate (1.6) SN/N = 0{\ogN), and this order- estimate is best possible. See Section [21 In Section [3] we will prove Theorem 1.3. Let the series Yl^n/n^ with coefficients a„ > converge to f{z) = f{x + iy) for x > 1. Setting sn = 'Yn<N ^n as before, the sequence {sj\f/N} will remain bounded if and only if the quotient (1.7) q{x + ty) = ^^^^ (x>l) X + ty converges in the sense of tempered distributions to a pseudomeasure q{l+iy) as X \ 1. A pseudomeasure is the distributional Fourier transform of a bounded measurable function. It has local representations by Fourier series with uniformly bounded coefficients. A simple example is given by the delta dis- tribution or Dirac measure. The following pseudomeasure is the boundary IKEHARA-TYPE THEOREM INVOLVING BOUNDEDNESS distribution of an analytic function: 1 1 /"CO '^'^ lim ^^ = lim / e-^*e-*^*rft. +0 + iy a;\o X + iy x\o Jq It is the Fourier transform of the Heaviside function l+(t), which equals 1 for t > and for t < 0. Pseudomeasures can have no singularities worse than first-order poles; cf. f l3.2p below. 2. Proof of Proposition 11.21 The proof consists of two parts. (i) Let f{z) = 'Y^ a-n/n^ with a„ > as in (11.11) satisfy the real condition (11. 5p . Setting x = xn = '^ + 1/logA^, one finds that ^-^ n ^-^ n n<N n<N A crude estimate now gives the result of (11.61) : N sj^ = y n—< NaN = O(NlogN). ^-^ n 1 (ii) For the second part we use an example. Lemma 2.1. Let 22*+fc forn = 2'^\ A; = 1, 2, • • • , """ ^ for all other n. (2.2) ■ = 0{l/{x - 1)} as x\l, hut (2.3) for N = 2^'' , one has Sn ^ cln = (1/ log 2) A^ log N. Proof. Takex = 1 + 5 with <6 <1. Then /(^) = 92^+fc ofc / ^ 22*'5 ■ Observe that the graph of h{t) = 2* ^ (0 < t < oo) 4 JACOB KOREVAAR is rising to a maximum at some point t = to (5) and then falling. Thus the sum for f{x) is majorized by the integral of h(t) over (0, oo) plus the value h(to). Both have the form const /S, hence ( 12.21) . Now take N of the form 2^*, so that log A^ = 2^ log 2. Then a^ = 22'+^= = N2'' = Ar(logAr)/log2. D 3. Proof of Theorem 11.31 Note that the distributional convergence in the theorem is convergence in the Schwartz space S'. In other words, (3.1) <q{x + iy),(j){y)> ^ <q{l + iy),(j){y)> as x \ 1 for all testing functions (f){y) G S, that is, all rapidly decreasing C°° func- tions; see Schwartz [9J or Hormander [3]. Proof of Theorem \1.3[ Let f{z) be the sum of the Dirichlet series in the theorem. Now define s{v) = J2n<v ^n, so that s(f) = for f < and stv = s{N) = 0{N^'^'^) for every e > 0. Integrating by parts, one obtains a representation for q{z) = f{z)/z as a Mellin transform: /CXD /»00 v~''ds{v) = / s{v)v^''^'^dv {x > 1). The substitution t> = e* gives q{z) as a shifted Laplace transform of S(t) = e~*s(e*): /•OO /•OO q{z) = / s{e')e~''dt = / S{t)e-^'-^^'dt {x > 1). Jo Jo (i) Suppose that the sequence {sn/N} is bounded. Then S{t) is bounded, \S{t)\ < M, say. Hence M (3.2) \q(z)\ < for x > 1. X — 1 Thus the boundary singularities of q{z) on the line {x = 1} can be no worse than first-order poles. We will verify that in the sense of distributions, q{x + iy) -^ g(l + iy) = S{y), where S{y) denotes the distributional Fourier transform of S{t). Indeed, for fixed x > 1, the function q{x + iy) is the Fourier transform of Sx{t) = 5(t)e-(^-i)*. Since \S{t)\ < M and S{t) = for t < 0, the functions IKEHARA-TYPE THEOREM INVOLVING BOUNDEDNESS 5 Sx{t) converge to S(t) boundedly as x \ 1, hence in the sense of tempered distributions. Since distributional Fourier transformation is continuous on the Schwartz space S', it follows that q{x+iy) converges to the distributional Fourier transform of S{t) - in this case a pseudomeasure. (ii) Conversely, suppose that q{x + iy) = Sx{y) converges to a pseudo- measure as X \ 1, symbolically written as g(l + iy). Then g(l + iy) is the Fourier transform H{y) of a bounded function H{t). By the continuity of inverse Fourier transformation, this implies that H{t) is the distributional limit of Sx{t) = S{t)e~^^~^^^ as x \ 1. But the latter limit is equal to S{t): /■oo < Sx{t), Mt) >= S.,{t)Mt)dt -^ < S{t), 0o(t) > Jo for all C°° functions 0o(i) of compact support. It follows that S(t) = H{t) on M, hence bounded. D 4. Final remarks Let 'iT2{N) denote the number of prime twins {p, p + 2) with p < N. The famous twin-prime conjecture (TPC) of Hardy and Littlewood [2] asserts that for N ^ oo. N N -'2: r'^ dt (4.1) 7:,{N) r. 2C,UN) = 2Cj ^20, J2 log t log A^ Here C2 is the 'twin-prime constant', C2= n {l-7^3Tp} ^0-6601618. p prime, p>2 '^ For the discussion of the TPC it is convenient to introduce the modified counting function (4.2) ^2(iV) = J] A(n)A(n + 2), n<N where A(fc) denotes von Mangoldt's function. Since A{k) = logp if fc = p" for some prime number p and A{k) = otherwise, the TPC turns out to be equivalent to the asymptotic relation (4.3) V'2(iV) ~ 2C2N as A^ ^ 00. 6 JACOB KOREVAAR It is natural then to introduce the Dirichlet series ,..-.. ^ / N def v^ A(n)A(n + 2) , 4.4 D2iz)''^y2^^ iz = x + ty,x>l). n=l By a sieving argument, cf. Halberstam and Richert [T], one has tt2{N) = OiN/log^N), or equivalents, ip2{N) = 0{N). By Theorem O another equivalent statement is that the quotient D2{x + iy) / {x + iy) converges dis- tributionally to a pseudomeasure as x \ 1. And finally, by the two-way Ikehara- Wiener theorem referred to in Section [H the TPC is equivalent to the conjecture that the difference D2{z) — 2C2/{z — 1) has local pseudo- function boundary behavior as x \ 1. [lo; [11 References H. Halberstam, and H.-E. Richert, Sieve methods. Academic Press, London, 1974. G. H. Hardy and J. E. Littlewood, Some problems of 'partitio numerorum'. Ill: On the expression of a number as a sum of primes. Acta Math. 44 (1923), 1-70. Hormander, L., The Analysis of Linear Partial Differential Operators, vol. 1. Grundl. math. Wiss. vol. 256, Springer, Berlin, 1983. S. Ikehara, An extension of Landau's theorem in the analytic theory of numbers, J. Math, and Phys. 10 (1931), 1-12. J. Korevaar, A century of complex Tauberian theory, Bull. Amer. Math. Soc. (N.S.) 39 (2002), 475-531. J. Korevaar, Tauberian Theory, a Century of Developments, Grundl. math. Wiss. vol. 329, Springer, Berlin, 2004. J. Korevaar, Distributional Wiener-Ikehara theorem and twin primes, Indag. Math. (N.S.) 16 (2005), 37-49. H. Mhaskar, A question about the Wiener-Ikehara theorem. In e-mail of April 14, 2008. Schwartz, L., Theorie des Distributions I, II. Hermann, Paris, 1966. (First edition 1950/51.) N. Wiener, Tauberian theorems, Ann. of Math. 33 (1932), 1-100. N. Wiener, N. The Fourier Integral and Certain of its Applications, Cambridge Univ. Press, Cambridge, 1933. KdV Institute of Mathematics, University of Amsterdam, Plantage Muidergracht 24, 1018 TV Amsterdam, Netherlands E-mail address: J . KorevaarOuva . nl